- AI Fire
- Posts
- 💰 xAI’s Billion-Dollar Grid Monster
In partnership with
Elon Musk’s xAI just powered up a monster, a single-site AI cluster that now uses more electricity than San Francisco. Let’s see what xAI’s trillion-token dreams look like.
What's on FIRE 🔥
LEARNING PARTNER AIRCAMPUS
Imagine having AI experts running your calendar, inbox, socials, and documents - all at once, all on autopilot.
On Wednesday, 21st January at 9AM join a select group of innovators who are redefining how work gets done with AI.
In this 3-hour AI Agent Mastery Session, we’ll show you step-by-step how to design, deploy, and automate AI systems that run entire workflows without you lifting a finger.
Whether it’s managing projects, handling customers, generating content, or analyzing data, these agents will work across every platform you use.
No code. No tech jargon. Just a proven blueprint for turning your ideas into powerful AI-driven results - faster than you ever thought possible.
AI INSIGHTS
Elon Musk’s xAI just activated Colossus 2, a single-site AI training cluster. It already draws more power than San Francisco. FYI, you’re looking at:
A 1 GW AI supercomputer, purpose‑built for training the Grok series
Running on dedicated grid lines, on‑site substations, and liquid cooling
So, Colossus 1 launched in about 122 days. Then, Colossus 2 followed just as fast, at even bigger scale. Everything happens in one location.
Global batch sizes are now limited by optimization, not hardware. It’s built for continuous trillion‑parameter training with zero compromises. To get here, xAI:
Bought 555,000+ GPUs, costing roughly $18B
Raised a $20B Series E, blowing past its $15B goal
Brought in Fidelity, QIA, Stepstone, Valor, MGX, Baron, plus partners NVIDIA and Cisco
Now, Colossus 1 + 2 now powers Grok 4, Grok Voice, Grok Imagine & the next flagship: Grok 5. To “understand the universe,” you need the world’s biggest brain. Colossus 2 is how xAI plans to build it.
PRESENTED BY ATTIO
Introducing the first AI-native CRM
Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.
With AI at the core, Attio lets you:
Prospect and route leads with research agents
Get real-time insights during customer calls
Build powerful automations for your complex workflows
Join industry leaders like Granola, Taskrabbit, Flatfile and more.
AI SOURCES FROM AI FIRE
🔥 Ep 42 Tooldrop: Builds a Zero-Effort Second Brain That Understands Your Work As Well As You Do
Today we explore and test 5 AI tools that creates long-form videos with consistent characters, clear story flow, and cinematic styles, all from a single prompt.
→ Get your full breakdown here (no hidden fee)!
1. Google’s new “Personal Intelligence” makes Gemini feel soo personal. It turns to a weirdly well-informed best friend who remembers everything about you
2. 95% tools are trash? I ranked 24 Best AIs to get Final Money-Making AI Stack! These 24 tools from S to F-Tier help you get real ROI, not more subscriptions
3. 7 Easiest AI Side Hustles that even students can start to get $1K/mo. Turn basic AI apps into money-making machines between my busy lectures today
PERFECTLY FORMATTED REPORTS INSTANTLY
You spend 3 hours prompting the perfect report, you hit "Paste" into Google Docs/Word, and suddenly... disaster!?
If you are spending more time re-formatting AI output or fixing fonts, spacing, headers, logos, or structure manually. This guide is a must-read.
You can get perfectly formatted documents that match your company’s exact styles. I promise your output will be:
Instantly Polished: Headers that actually look like headers (H1, H2), not just "big bold text"
Perfectly Structured: Bullet points that indent correctly and don't turn into random dashes
Brand Ready: Documents that land in your editor matching your company’s style guide, with zero manual tweaks
TODAY IN AI
AI HIGHLIGHTS
🛠️ If you're “vibe coding” with Claude and not using these 200+ sub-agent prompts, you're missing a fun list to make your flow smoother, and more modular.
🌚 Google just says no ads in Gemini (yet), while OpenAI is testing ads inside GPT. Google VP explained why, read it & you'll feel who’s feeling the pressure for real.
📈 Gemini API traffic more than doubled in just 5 months from 35B to 85B requests (mostly from Gemini 2.5 and 3 driving the surge). It’s now even profitable on ops.
😮 While AI eats the internet, people are literally knitting. Yarn kit sales at Michael’s jumped 1,200%, “analog lifestyle” searches exploded here. Why 2026 feels like 1996?
📈 OpenAI just revealed $20B+ annualized revenue for 2025, tripling year over year, with 10x compute expansion since 2023. CFO Sarah shared all in a new blog post.
💰 Big AI Fundraising: Alibaba‑backed Moonshot AI is now valued at $4.8B, up $500M from last month, as Chinese AI rivals like MiniMax surge after Hong Kong IPOs.
NEW EMPOWERED AI TOOLS
🍜 Noodle Seed connects your business to AI conversations like ChatGPT, Gemini, Claude, Perplexity,… in minutes
👤 OS Ninja turns complex code repositories into interactive, AI-guided learning journeys for you
✨ FlowGenie makes building forms and automating workflows feel like magic, using a visual flow editor
🗺️ Tofu Maps lets you create personal maps, save places, and share them by link without signing up
AI BREAKTHROUGH
If you’ve ever wondered how Google plans to actually monitor AI misuse without lighting money on fire, this new DeepMind paper has the answer.
They’re calling them activation probes, like tiny classifiers that scan the model’s brain as it answers a prompt to flag suspicious activity. Compared to full LLM monitors, probes are:
10,000x cheaper for broad traffic scanning
Still accurate (especially when tuned for harmful snippets in long contexts)
Already used in Gemini deployments
All the probe does is a small matrix read on already-computed data. It’s like eavesdropping on the model’s thoughts before it speaks, and it's nearly free. Not just one probe, it’s actually a stack of smart designs:
MultiMax probes pick out strong local signals in long prompts (even up to 1M tokens)
Rolling Mean Attention filters out noise and spots brief toxic spans
Monitoring AI misuse is hard. LLM-based safety systems cost money, tokens, and time. You can’t run “Is this safe?” on every single user prompt at scale. So this changes that.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply